A team has no reliable idea of what normal request latency looks like across the day and does not want to hand-tune static thresholds. Which CloudWatch feature should it use?
Show answer and explanation
Answer: A
Anomaly detection trains a model on a metric's history, including its daily and weekly patterns, and produces a band that represents expected values. An alarm then fires when the metric moves outside that band for the configured number of evaluation periods.
The band adapts as the workload's baseline changes, so seasonal spikes do not trigger constant noise the way a fixed threshold would.
Why the other options are wrong
- B. Composite alarms combine existing alarm states and still depend on the static thresholds you were trying to avoid.
- C. Metric math derives new time series from existing metrics but does not learn a baseline or create a band.
- D. Dashboards display data but never evaluate it or notify anyone.
Exam tip. Anomaly detection builds the band for you; the alarm just watches it.